When I was first being introduced to Effective Altruism, I was told that its central contribution to the world was the “Importance, Neglectedness, Tractability” (INT) heuristic. However, having now inhabited EA-land for a few years, I’m no longer sure that this heuristic is especially meaningful (or even very widely used), and I would advocate for removing it from promotional materials.
The nugget of what EA is really “all about” is really buried somewhere inside “Importance.” Something like what is important about an action is how people feel as a result of the action. Or that the world is a morally important place because it contains people who can be made better or worse off. Or maybe it’s easiest to stick with: “the greatest good for the greatest number.”
Anyway, in my personal work, I’ve found it most useful to use the following fraction:
(Shouldn’t it be subjects times average experience? Wouldn’t this fraction imply that better experiences should be assigned low numbers? Isn’t that slightly awkward? Er, probably. But in my head this analysis involves two distinct steps: the numerator, which is proudly inflationary, and the denominator, which is pessimistic and conservative. So it’s mentally convenient for me to put them on opposite sides of the vinculum.)
First, you give free rein to your imagination and come up with the biggest number you possibly can. A mental health intervention in Africa won’t just make the mother happier; it’ll help her kids get fed and educated, which will improve the whole village’s economy. A cheap cultivated steak won’t just be another option in the supermarket; it’ll present an alternate moral vision to billions of people. Saving the world now won’t just let us live another day; it’ll allow us to colonize the galaxy and manufacture digital minds. This step is a lot of fun.
Then you stop to consider just how much better off those subjects will really be. This step is sort of dour. So the African kids get educated — so what? How many extra DALYs will they really receive from their education? How much would they prefer those years to an early death? Will they spend the extra years eating meat and fish? Will they care enough to choose the cultivated kind, even if it’s cheaper and tastes just as good? Will they have more children? Will their descendants care enough to choose cultivated meat? Will their descendants care about the wellbeing of future digital minds? Will future digital minds care about the wellbeing of future humans? Will future digital minds even care about each other’s wellbeing? How much better is all this, exactly, then buying yourself (or some other equivalently middle-income schmuck) a bitching new car?
I think that EA is structurally better positioned to fill in the numerator than the denominator.
The numerator mostly involves physical facts and a healthy helping of creative techno-optimism. For instance: depression is a neurological phenomenon; animals are inefficient at transforming solar energy into edible calories; stellar radiation provides significantly more energy than our brains currently consume. Ergo: deep brain stimulation, cell-cultivated meat, and digital minds powered by a Dyson sphere. The numerator is about what is possible.
The denominator is murkier and more introspective. It definitely requires propositions about the subjective value of different experiences, which are always difficult. You are definitely better off than a factory-farmed chicken, but by how much? Sometimes it’s helpful to frame these questions as gambles, but it’s not that much more helpful, e.g. Would you rather have a 1/200,000,000,000 chance of being reborn as a factory-farmed chicken, or a ½ chance of being reborn as a slightly unhappy Canadian? (For that matter, would you like to be reborn as anyone at all? For me, the answer changes from day to day.)
I feel wildly clueless about whether it is 100,000,000,000 or 100,000,000,000,000,000,000,000 times worse to be a factory-farmed chicken than a sad Canadian, a la Anthony DiGiovanni’s oscillating scales. Perhaps I could narrow the gap by meditating more on my own desires, or fine-tuning my ability to distinguish between miniscule probabilities, but I suspect that there will remain a certain hard kernel of cluelessness between my upper and lower bounds, probably spanning many orders of magnitude. This variety of cluelessness is what I will call the denominator problem.
Critics sometimes present the denominator problem as the movement’s Achilles Heel — its lack of guidance about what people should want. I am fairly sure that this is a bad-faith objection, since alternative proscriptions for what should and should not be wanted always seem preposterous and insane. And either way, most non-EAs (including the critics) don’t generally seem to act in a way that is internally consistent with their own ethics, as far as I can tell.
All the same, I am occasionally annoyed and kept awake by EA’s silence on the denominator problem, especially contrasted with its enthusiastic stabs at the numerator. Many EAs seem to bracket out the denominator entirely when they are choosing between interventions. Hence the common progression of concerns as people learn more about EA:
Humans -> Farmed animals -> Wild animals -> Future humans/digital minds
It is surely true that there are more future beings than wild animals, more wild animals than farmed animals, and more farmed animals than humans. 10^54 is way bigger than 10^20, which is a lot bigger than 10^11, which is significantly bigger than 10^9. (I even made a log-scaled interactive tool for adding up different populations, in order to understand what a mind-blowingly small rounding error some of them are relative to each other. Only a few farmed animal populations even remain visible when you stack them on each other.) The steps in this progression often contain substeps, like the shift from strict vegan leafleting to cage-free campaigns to shrimp stunning.
But these kinds of sequences are entirely driven by the numerator. While there are more shrimp than chickens, and more chickens than cows, there’s no reason to think that there aren’t also 45 orders of magnitude between stunned slaughter, cage-free factory farmed-ness, and nonexistence.
EAs do sometimes harbor denominator-related concerns. A classic: do farmed animals really feel pain as intensely as a human does? And occasionally: are wild animals really that unhappy? But by the time an EA has ridden the crazy train all the way to longtermism, she has usually learned to stay quiet on the denominator question, and not to wonder about how happy future digital minds will really be relative to humans and animals who are suffering now.
Partly this is a practical instinct. For instance, while I admire Rethink Priorities’ cleverness in assembling the Moral Weights Project, several critics have usefully pointed out that it answered a subtly different question from the denominator problem. We might someday be able to conduct experiments where we repeatedly ask humans (hopefully volunteers) how much they would like to pay (if anything) to have their brain be switched out for a pig’s brain right before they undergo a painful experience. (After a few trials, maybe they would become willing to pay not to have the procedure — that would be interesting). If we did this on a sufficient number of humans, we might come up with a workable average conversion rate.
Yet even then we might be haunted by the possibility that the subjects were misrepresenting their experience, or that the experiment had been polluted by their distaste for the mechanics of the pig-brain transplant. In other words, the denominator problem keeps kicking for a while before it really dies. I think that the specter of these high philosophical standards, especially contrasted with gleeful techno-optimist numerator projections, quietly pushes EAs to abandon the denominator entirely and stick to the numerator as a reasonable proxy.
For my part, I am happy to abandon the denominator in the case of farmed animals, since I have a strong conviction that animals with similar neural anatomy and pain responses to my own are suffering just as much as I am. I am less happy to abandon it in the case of wild animals, since I have confused intuitions about whether their lives are positive or negative. I am also slightly unwilling to abandon it for future beings, since I have some conviction that the presence of self-interest means that future beings (both human and digital) will never reach any kind of unanimous blissful heaven — that some population will still end up hurt and ignored, and that the presence of those populations may make the future a morally dubious reward for forgoing goods in the present.
These intuitions are obviously weak, especially judged against the standards of the pig-brain transplant. But the answer can’t be that we should ignore the denominator entirely.
In conclusion, I would like to submit the denominator for special consideration as the Forum continues cogitating on the nature of cluelessness. In the meantime, I’d like to urge everyone to take the denominator more seriously. Watch footage of factory farms: try to imagine that you were feeling the intensity of suffering felt by a single chicken, and consider exactly how much better you would feel if you were let out of the cage, but kept in the same barn. The same? 10 times better? 100,000,000,000 times better? Or read history and sociological experiments and try to determine how future human civilizations will treat their digital minds. The same as the chicken? 10 times better? 100,000,000,000 times better?
These are unnatural questions, probably lacking good answers, but it seems perilous to ignore them — they’re at least half the ballgame.